Evaluating Random Forest Model Performance in Predicting Photovoltaic Power Generation Across Puerto Rico
Open Access
Author:
Vitti, Caden
Area of Honors:
Energy Engineering
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Renee Obringer, Thesis Supervisor Nelson Yaw Dzade, Thesis Honors Advisor
Keywords:
Puerto Rico resilience microgrid random forest solar power climate change
Abstract:
As the climate crisis intensifies and more extreme weather events hit vulnerable communities around the world, it is becoming increasingly important to conduct further research on energy systems and their relationships with climate change, described as the renewables-climate nexus. In particular, the utilization of microgrids to increase resilience for grid-scale or local-scale energy networks is a growing area of focus. In the case of Puerto Rico, which has suffered from a weak energy infrastructure and multiple hurricanes, a specialized model that incorporates climate data into renewable microgrid generation prediction could provide opportunities for grassroots development. Here, I present a random forest regression machine learning model that utilizes climate data from 48 stations across 6 regions in Puerto Rico to predict power generation values for 3 different types of solar panels. The results of this study present the model as successful with low error values and strong relationships between nonlinear predictor and response variables. Despite socio-geographical differences across the regions of the island, this model creates a baseline which developers can use when factoring resiliency into planning of future microgrid networks.
